aboutsummaryrefslogtreecommitdiffstats
path: root/virt/kvm/kvm_main.c
diff options
context:
space:
mode:
authorBenjamin Herrenschmidt <benh@kernel.crashing.org>2009-12-09 01:14:38 -0500
committerBenjamin Herrenschmidt <benh@kernel.crashing.org>2009-12-09 01:14:38 -0500
commitbcd6acd51f3d4d1ada201e9bc5c40a31d6d80c71 (patch)
tree2f6dffd2d3e4dd67355a224de7e7a960335a92fd /virt/kvm/kvm_main.c
parent11c34c7deaeeebcee342cbc35e1bb2a6711b2431 (diff)
parent3ff6a468b45b5dfeb0e903e56f4eb27d34b2437c (diff)
Merge commit 'origin/master' into next
Conflicts: include/linux/kvm.h
Diffstat (limited to 'virt/kvm/kvm_main.c')
-rw-r--r--virt/kvm/kvm_main.c961
1 files changed, 147 insertions, 814 deletions
diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index cdca63917e77..e1f2bf8d7b1e 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -43,6 +43,7 @@
43#include <linux/swap.h> 43#include <linux/swap.h>
44#include <linux/bitops.h> 44#include <linux/bitops.h>
45#include <linux/spinlock.h> 45#include <linux/spinlock.h>
46#include <linux/compat.h>
46 47
47#include <asm/processor.h> 48#include <asm/processor.h>
48#include <asm/io.h> 49#include <asm/io.h>
@@ -54,12 +55,6 @@
54#include "coalesced_mmio.h" 55#include "coalesced_mmio.h"
55#endif 56#endif
56 57
57#ifdef KVM_CAP_DEVICE_ASSIGNMENT
58#include <linux/pci.h>
59#include <linux/interrupt.h>
60#include "irq.h"
61#endif
62
63#define CREATE_TRACE_POINTS 58#define CREATE_TRACE_POINTS
64#include <trace/events/kvm.h> 59#include <trace/events/kvm.h>
65 60
@@ -76,6 +71,8 @@ DEFINE_SPINLOCK(kvm_lock);
76LIST_HEAD(vm_list); 71LIST_HEAD(vm_list);
77 72
78static cpumask_var_t cpus_hardware_enabled; 73static cpumask_var_t cpus_hardware_enabled;
74static int kvm_usage_count = 0;
75static atomic_t hardware_enable_failed;
79 76
80struct kmem_cache *kvm_vcpu_cache; 77struct kmem_cache *kvm_vcpu_cache;
81EXPORT_SYMBOL_GPL(kvm_vcpu_cache); 78EXPORT_SYMBOL_GPL(kvm_vcpu_cache);
@@ -86,615 +83,13 @@ struct dentry *kvm_debugfs_dir;
86 83
87static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl, 84static long kvm_vcpu_ioctl(struct file *file, unsigned int ioctl,
88 unsigned long arg); 85 unsigned long arg);
86static int hardware_enable_all(void);
87static void hardware_disable_all(void);
89 88
90static bool kvm_rebooting; 89static bool kvm_rebooting;
91 90
92static bool largepages_enabled = true; 91static bool largepages_enabled = true;
93 92
94#ifdef KVM_CAP_DEVICE_ASSIGNMENT
95static struct kvm_assigned_dev_kernel *kvm_find_assigned_dev(struct list_head *head,
96 int assigned_dev_id)
97{
98 struct list_head *ptr;
99 struct kvm_assigned_dev_kernel *match;
100
101 list_for_each(ptr, head) {
102 match = list_entry(ptr, struct kvm_assigned_dev_kernel, list);
103 if (match->assigned_dev_id == assigned_dev_id)
104 return match;
105 }
106 return NULL;
107}
108
109static int find_index_from_host_irq(struct kvm_assigned_dev_kernel
110 *assigned_dev, int irq)
111{
112 int i, index;
113 struct msix_entry *host_msix_entries;
114
115 host_msix_entries = assigned_dev->host_msix_entries;
116
117 index = -1;
118 for (i = 0; i < assigned_dev->entries_nr; i++)
119 if (irq == host_msix_entries[i].vector) {
120 index = i;
121 break;
122 }
123 if (index < 0) {
124 printk(KERN_WARNING "Fail to find correlated MSI-X entry!\n");
125 return 0;
126 }
127
128 return index;
129}
130
131static void kvm_assigned_dev_interrupt_work_handler(struct work_struct *work)
132{
133 struct kvm_assigned_dev_kernel *assigned_dev;
134 struct kvm *kvm;
135 int i;
136
137 assigned_dev = container_of(work, struct kvm_assigned_dev_kernel,
138 interrupt_work);
139 kvm = assigned_dev->kvm;
140
141 mutex_lock(&kvm->irq_lock);
142 spin_lock_irq(&assigned_dev->assigned_dev_lock);
143 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
144 struct kvm_guest_msix_entry *guest_entries =
145 assigned_dev->guest_msix_entries;
146 for (i = 0; i < assigned_dev->entries_nr; i++) {
147 if (!(guest_entries[i].flags &
148 KVM_ASSIGNED_MSIX_PENDING))
149 continue;
150 guest_entries[i].flags &= ~KVM_ASSIGNED_MSIX_PENDING;
151 kvm_set_irq(assigned_dev->kvm,
152 assigned_dev->irq_source_id,
153 guest_entries[i].vector, 1);
154 }
155 } else
156 kvm_set_irq(assigned_dev->kvm, assigned_dev->irq_source_id,
157 assigned_dev->guest_irq, 1);
158
159 spin_unlock_irq(&assigned_dev->assigned_dev_lock);
160 mutex_unlock(&assigned_dev->kvm->irq_lock);
161}
162
163static irqreturn_t kvm_assigned_dev_intr(int irq, void *dev_id)
164{
165 unsigned long flags;
166 struct kvm_assigned_dev_kernel *assigned_dev =
167 (struct kvm_assigned_dev_kernel *) dev_id;
168
169 spin_lock_irqsave(&assigned_dev->assigned_dev_lock, flags);
170 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
171 int index = find_index_from_host_irq(assigned_dev, irq);
172 if (index < 0)
173 goto out;
174 assigned_dev->guest_msix_entries[index].flags |=
175 KVM_ASSIGNED_MSIX_PENDING;
176 }
177
178 schedule_work(&assigned_dev->interrupt_work);
179
180 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_GUEST_INTX) {
181 disable_irq_nosync(irq);
182 assigned_dev->host_irq_disabled = true;
183 }
184
185out:
186 spin_unlock_irqrestore(&assigned_dev->assigned_dev_lock, flags);
187 return IRQ_HANDLED;
188}
189
190/* Ack the irq line for an assigned device */
191static void kvm_assigned_dev_ack_irq(struct kvm_irq_ack_notifier *kian)
192{
193 struct kvm_assigned_dev_kernel *dev;
194 unsigned long flags;
195
196 if (kian->gsi == -1)
197 return;
198
199 dev = container_of(kian, struct kvm_assigned_dev_kernel,
200 ack_notifier);
201
202 kvm_set_irq(dev->kvm, dev->irq_source_id, dev->guest_irq, 0);
203
204 /* The guest irq may be shared so this ack may be
205 * from another device.
206 */
207 spin_lock_irqsave(&dev->assigned_dev_lock, flags);
208 if (dev->host_irq_disabled) {
209 enable_irq(dev->host_irq);
210 dev->host_irq_disabled = false;
211 }
212 spin_unlock_irqrestore(&dev->assigned_dev_lock, flags);
213}
214
215static void deassign_guest_irq(struct kvm *kvm,
216 struct kvm_assigned_dev_kernel *assigned_dev)
217{
218 kvm_unregister_irq_ack_notifier(kvm, &assigned_dev->ack_notifier);
219 assigned_dev->ack_notifier.gsi = -1;
220
221 if (assigned_dev->irq_source_id != -1)
222 kvm_free_irq_source_id(kvm, assigned_dev->irq_source_id);
223 assigned_dev->irq_source_id = -1;
224 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_GUEST_MASK);
225}
226
227/* The function implicit hold kvm->lock mutex due to cancel_work_sync() */
228static void deassign_host_irq(struct kvm *kvm,
229 struct kvm_assigned_dev_kernel *assigned_dev)
230{
231 /*
232 * In kvm_free_device_irq, cancel_work_sync return true if:
233 * 1. work is scheduled, and then cancelled.
234 * 2. work callback is executed.
235 *
236 * The first one ensured that the irq is disabled and no more events
237 * would happen. But for the second one, the irq may be enabled (e.g.
238 * for MSI). So we disable irq here to prevent further events.
239 *
240 * Notice this maybe result in nested disable if the interrupt type is
241 * INTx, but it's OK for we are going to free it.
242 *
243 * If this function is a part of VM destroy, please ensure that till
244 * now, the kvm state is still legal for probably we also have to wait
245 * interrupt_work done.
246 */
247 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSIX) {
248 int i;
249 for (i = 0; i < assigned_dev->entries_nr; i++)
250 disable_irq_nosync(assigned_dev->
251 host_msix_entries[i].vector);
252
253 cancel_work_sync(&assigned_dev->interrupt_work);
254
255 for (i = 0; i < assigned_dev->entries_nr; i++)
256 free_irq(assigned_dev->host_msix_entries[i].vector,
257 (void *)assigned_dev);
258
259 assigned_dev->entries_nr = 0;
260 kfree(assigned_dev->host_msix_entries);
261 kfree(assigned_dev->guest_msix_entries);
262 pci_disable_msix(assigned_dev->dev);
263 } else {
264 /* Deal with MSI and INTx */
265 disable_irq_nosync(assigned_dev->host_irq);
266 cancel_work_sync(&assigned_dev->interrupt_work);
267
268 free_irq(assigned_dev->host_irq, (void *)assigned_dev);
269
270 if (assigned_dev->irq_requested_type & KVM_DEV_IRQ_HOST_MSI)
271 pci_disable_msi(assigned_dev->dev);
272 }
273
274 assigned_dev->irq_requested_type &= ~(KVM_DEV_IRQ_HOST_MASK);
275}
276
277static int kvm_deassign_irq(struct kvm *kvm,
278 struct kvm_assigned_dev_kernel *assigned_dev,
279 unsigned long irq_requested_type)
280{
281 unsigned long guest_irq_type, host_irq_type;
282
283 if (!irqchip_in_kernel(kvm))
284 return -EINVAL;
285 /* no irq assignment to deassign */
286 if (!assigned_dev->irq_requested_type)
287 return -ENXIO;
288
289 host_irq_type = irq_requested_type & KVM_DEV_IRQ_HOST_MASK;
290 guest_irq_type = irq_requested_type & KVM_DEV_IRQ_GUEST_MASK;
291
292 if (host_irq_type)
293 deassign_host_irq(kvm, assigned_dev);
294 if (guest_irq_type)
295 deassign_guest_irq(kvm, assigned_dev);
296
297 return 0;
298}
299
300static void kvm_free_assigned_irq(struct kvm *kvm,
301 struct kvm_assigned_dev_kernel *assigned_dev)
302{
303 kvm_deassign_irq(kvm, assigned_dev, assigned_dev->irq_requested_type);
304}
305
306static void kvm_free_assigned_device(struct kvm *kvm,
307 struct kvm_assigned_dev_kernel
308 *assigned_dev)
309{
310 kvm_free_assigned_irq(kvm, assigned_dev);
311
312 pci_reset_function(assigned_dev->dev);
313
314 pci_release_regions(assigned_dev->dev);
315 pci_disable_device(assigned_dev->dev);
316 pci_dev_put(assigned_dev->dev);
317
318 list_del(&assigned_dev->list);
319 kfree(assigned_dev);
320}
321
322void kvm_free_all_assigned_devices(struct kvm *kvm)
323{
324 struct list_head *ptr, *ptr2;
325 struct kvm_assigned_dev_kernel *assigned_dev;
326
327 list_for_each_safe(ptr, ptr2, &kvm->arch.assigned_dev_head) {
328 assigned_dev = list_entry(ptr,
329 struct kvm_assigned_dev_kernel,
330 list);
331
332 kvm_free_assigned_device(kvm, assigned_dev);
333 }
334}
335
336static int assigned_device_enable_host_intx(struct kvm *kvm,
337 struct kvm_assigned_dev_kernel *dev)
338{
339 dev->host_irq = dev->dev->irq;
340 /* Even though this is PCI, we don't want to use shared
341 * interrupts. Sharing host devices with guest-assigned devices
342 * on the same interrupt line is not a happy situation: there
343 * are going to be long delays in accepting, acking, etc.
344 */
345 if (request_irq(dev->host_irq, kvm_assigned_dev_intr,
346 0, "kvm_assigned_intx_device", (void *)dev))
347 return -EIO;
348 return 0;
349}
350
351#ifdef __KVM_HAVE_MSI
352static int assigned_device_enable_host_msi(struct kvm *kvm,
353 struct kvm_assigned_dev_kernel *dev)
354{
355 int r;
356
357 if (!dev->dev->msi_enabled) {
358 r = pci_enable_msi(dev->dev);
359 if (r)
360 return r;
361 }
362
363 dev->host_irq = dev->dev->irq;
364 if (request_irq(dev->host_irq, kvm_assigned_dev_intr, 0,
365 "kvm_assigned_msi_device", (void *)dev)) {
366 pci_disable_msi(dev->dev);
367 return -EIO;
368 }
369
370 return 0;
371}
372#endif
373
374#ifdef __KVM_HAVE_MSIX
375static int assigned_device_enable_host_msix(struct kvm *kvm,
376 struct kvm_assigned_dev_kernel *dev)
377{
378 int i, r = -EINVAL;
379
380 /* host_msix_entries and guest_msix_entries should have been
381 * initialized */
382 if (dev->entries_nr == 0)
383 return r;
384
385 r = pci_enable_msix(dev->dev, dev->host_msix_entries, dev->entries_nr);
386 if (r)
387 return r;
388
389 for (i = 0; i < dev->entries_nr; i++) {
390 r = request_irq(dev->host_msix_entries[i].vector,
391 kvm_assigned_dev_intr, 0,
392 "kvm_assigned_msix_device",
393 (void *)dev);
394 /* FIXME: free requested_irq's on failure */
395 if (r)
396 return r;
397 }
398
399 return 0;
400}
401
402#endif
403
404static int assigned_device_enable_guest_intx(struct kvm *kvm,
405 struct kvm_assigned_dev_kernel *dev,
406 struct kvm_assigned_irq *irq)
407{
408 dev->guest_irq = irq->guest_irq;
409 dev->ack_notifier.gsi = irq->guest_irq;
410 return 0;
411}
412
413#ifdef __KVM_HAVE_MSI
414static int assigned_device_enable_guest_msi(struct kvm *kvm,
415 struct kvm_assigned_dev_kernel *dev,
416 struct kvm_assigned_irq *irq)
417{
418 dev->guest_irq = irq->guest_irq;
419 dev->ack_notifier.gsi = -1;
420 dev->host_irq_disabled = false;
421 return 0;
422}
423#endif
424#ifdef __KVM_HAVE_MSIX
425static int assigned_device_enable_guest_msix(struct kvm *kvm,
426 struct kvm_assigned_dev_kernel *dev,
427 struct kvm_assigned_irq *irq)
428{
429 dev->guest_irq = irq->guest_irq;
430 dev->ack_notifier.gsi = -1;
431 dev->host_irq_disabled = false;
432 return 0;
433}
434#endif
435
436static int assign_host_irq(struct kvm *kvm,
437 struct kvm_assigned_dev_kernel *dev,
438 __u32 host_irq_type)
439{
440 int r = -EEXIST;
441
442 if (dev->irq_requested_type & KVM_DEV_IRQ_HOST_MASK)
443 return r;
444
445 switch (host_irq_type) {
446 case KVM_DEV_IRQ_HOST_INTX:
447 r = assigned_device_enable_host_intx(kvm, dev);
448 break;
449#ifdef __KVM_HAVE_MSI
450 case KVM_DEV_IRQ_HOST_MSI:
451 r = assigned_device_enable_host_msi(kvm, dev);
452 break;
453#endif
454#ifdef __KVM_HAVE_MSIX
455 case KVM_DEV_IRQ_HOST_MSIX:
456 r = assigned_device_enable_host_msix(kvm, dev);
457 break;
458#endif
459 default:
460 r = -EINVAL;
461 }
462
463 if (!r)
464 dev->irq_requested_type |= host_irq_type;
465
466 return r;
467}
468
469static int assign_guest_irq(struct kvm *kvm,
470 struct kvm_assigned_dev_kernel *dev,
471 struct kvm_assigned_irq *irq,
472 unsigned long guest_irq_type)
473{
474 int id;
475 int r = -EEXIST;
476
477 if (dev->irq_requested_type & KVM_DEV_IRQ_GUEST_MASK)
478 return r;
479
480 id = kvm_request_irq_source_id(kvm);
481 if (id < 0)
482 return id;
483
484 dev->irq_source_id = id;
485
486 switch (guest_irq_type) {
487 case KVM_DEV_IRQ_GUEST_INTX:
488 r = assigned_device_enable_guest_intx(kvm, dev, irq);
489 break;
490#ifdef __KVM_HAVE_MSI
491 case KVM_DEV_IRQ_GUEST_MSI:
492 r = assigned_device_enable_guest_msi(kvm, dev, irq);
493 break;
494#endif
495#ifdef __KVM_HAVE_MSIX
496 case KVM_DEV_IRQ_GUEST_MSIX:
497 r = assigned_device_enable_guest_msix(kvm, dev, irq);
498 break;
499#endif
500 default:
501 r = -EINVAL;
502 }
503
504 if (!r) {
505 dev->irq_requested_type |= guest_irq_type;
506 kvm_register_irq_ack_notifier(kvm, &dev->ack_notifier);
507 } else
508 kvm_free_irq_source_id(kvm, dev->irq_source_id);
509
510 return r;
511}
512
513/* TODO Deal with KVM_DEV_IRQ_ASSIGNED_MASK_MSIX */
514static int kvm_vm_ioctl_assign_irq(struct kvm *kvm,
515 struct kvm_assigned_irq *assigned_irq)
516{
517 int r = -EINVAL;
518 struct kvm_assigned_dev_kernel *match;
519 unsigned long host_irq_type, guest_irq_type;
520
521 if (!capable(CAP_SYS_RAWIO))
522 return -EPERM;
523
524 if (!irqchip_in_kernel(kvm))
525 return r;
526
527 mutex_lock(&kvm->lock);
528 r = -ENODEV;
529 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
530 assigned_irq->assigned_dev_id);
531 if (!match)
532 goto out;
533
534 host_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_HOST_MASK);
535 guest_irq_type = (assigned_irq->flags & KVM_DEV_IRQ_GUEST_MASK);
536
537 r = -EINVAL;
538 /* can only assign one type at a time */
539 if (hweight_long(host_irq_type) > 1)
540 goto out;
541 if (hweight_long(guest_irq_type) > 1)
542 goto out;
543 if (host_irq_type == 0 && guest_irq_type == 0)
544 goto out;
545
546 r = 0;
547 if (host_irq_type)
548 r = assign_host_irq(kvm, match, host_irq_type);
549 if (r)
550 goto out;
551
552 if (guest_irq_type)
553 r = assign_guest_irq(kvm, match, assigned_irq, guest_irq_type);
554out:
555 mutex_unlock(&kvm->lock);
556 return r;
557}
558
559static int kvm_vm_ioctl_deassign_dev_irq(struct kvm *kvm,
560 struct kvm_assigned_irq
561 *assigned_irq)
562{
563 int r = -ENODEV;
564 struct kvm_assigned_dev_kernel *match;
565
566 mutex_lock(&kvm->lock);
567
568 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
569 assigned_irq->assigned_dev_id);
570 if (!match)
571 goto out;
572
573 r = kvm_deassign_irq(kvm, match, assigned_irq->flags);
574out:
575 mutex_unlock(&kvm->lock);
576 return r;
577}
578
579static int kvm_vm_ioctl_assign_device(struct kvm *kvm,
580 struct kvm_assigned_pci_dev *assigned_dev)
581{
582 int r = 0;
583 struct kvm_assigned_dev_kernel *match;
584 struct pci_dev *dev;
585
586 down_read(&kvm->slots_lock);
587 mutex_lock(&kvm->lock);
588
589 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
590 assigned_dev->assigned_dev_id);
591 if (match) {
592 /* device already assigned */
593 r = -EEXIST;
594 goto out;
595 }
596
597 match = kzalloc(sizeof(struct kvm_assigned_dev_kernel), GFP_KERNEL);
598 if (match == NULL) {
599 printk(KERN_INFO "%s: Couldn't allocate memory\n",
600 __func__);
601 r = -ENOMEM;
602 goto out;
603 }
604 dev = pci_get_bus_and_slot(assigned_dev->busnr,
605 assigned_dev->devfn);
606 if (!dev) {
607 printk(KERN_INFO "%s: host device not found\n", __func__);
608 r = -EINVAL;
609 goto out_free;
610 }
611 if (pci_enable_device(dev)) {
612 printk(KERN_INFO "%s: Could not enable PCI device\n", __func__);
613 r = -EBUSY;
614 goto out_put;
615 }
616 r = pci_request_regions(dev, "kvm_assigned_device");
617 if (r) {
618 printk(KERN_INFO "%s: Could not get access to device regions\n",
619 __func__);
620 goto out_disable;
621 }
622
623 pci_reset_function(dev);
624
625 match->assigned_dev_id = assigned_dev->assigned_dev_id;
626 match->host_busnr = assigned_dev->busnr;
627 match->host_devfn = assigned_dev->devfn;
628 match->flags = assigned_dev->flags;
629 match->dev = dev;
630 spin_lock_init(&match->assigned_dev_lock);
631 match->irq_source_id = -1;
632 match->kvm = kvm;
633 match->ack_notifier.irq_acked = kvm_assigned_dev_ack_irq;
634 INIT_WORK(&match->interrupt_work,
635 kvm_assigned_dev_interrupt_work_handler);
636
637 list_add(&match->list, &kvm->arch.assigned_dev_head);
638
639 if (assigned_dev->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU) {
640 if (!kvm->arch.iommu_domain) {
641 r = kvm_iommu_map_guest(kvm);
642 if (r)
643 goto out_list_del;
644 }
645 r = kvm_assign_device(kvm, match);
646 if (r)
647 goto out_list_del;
648 }
649
650out:
651 mutex_unlock(&kvm->lock);
652 up_read(&kvm->slots_lock);
653 return r;
654out_list_del:
655 list_del(&match->list);
656 pci_release_regions(dev);
657out_disable:
658 pci_disable_device(dev);
659out_put:
660 pci_dev_put(dev);
661out_free:
662 kfree(match);
663 mutex_unlock(&kvm->lock);
664 up_read(&kvm->slots_lock);
665 return r;
666}
667#endif
668
669#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
670static int kvm_vm_ioctl_deassign_device(struct kvm *kvm,
671 struct kvm_assigned_pci_dev *assigned_dev)
672{
673 int r = 0;
674 struct kvm_assigned_dev_kernel *match;
675
676 mutex_lock(&kvm->lock);
677
678 match = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
679 assigned_dev->assigned_dev_id);
680 if (!match) {
681 printk(KERN_INFO "%s: device hasn't been assigned before, "
682 "so cannot be deassigned\n", __func__);
683 r = -EINVAL;
684 goto out;
685 }
686
687 if (match->flags & KVM_DEV_ASSIGN_ENABLE_IOMMU)
688 kvm_deassign_device(kvm, match);
689
690 kvm_free_assigned_device(kvm, match);
691
692out:
693 mutex_unlock(&kvm->lock);
694 return r;
695}
696#endif
697
698inline int kvm_is_mmio_pfn(pfn_t pfn) 93inline int kvm_is_mmio_pfn(pfn_t pfn)
699{ 94{
700 if (pfn_valid(pfn)) { 95 if (pfn_valid(pfn)) {
@@ -950,6 +345,7 @@ static const struct mmu_notifier_ops kvm_mmu_notifier_ops = {
950 345
951static struct kvm *kvm_create_vm(void) 346static struct kvm *kvm_create_vm(void)
952{ 347{
348 int r = 0;
953 struct kvm *kvm = kvm_arch_create_vm(); 349 struct kvm *kvm = kvm_arch_create_vm();
954#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 350#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
955 struct page *page; 351 struct page *page;
@@ -957,16 +353,21 @@ static struct kvm *kvm_create_vm(void)
957 353
958 if (IS_ERR(kvm)) 354 if (IS_ERR(kvm))
959 goto out; 355 goto out;
356
357 r = hardware_enable_all();
358 if (r)
359 goto out_err_nodisable;
360
960#ifdef CONFIG_HAVE_KVM_IRQCHIP 361#ifdef CONFIG_HAVE_KVM_IRQCHIP
961 INIT_LIST_HEAD(&kvm->irq_routing);
962 INIT_HLIST_HEAD(&kvm->mask_notifier_list); 362 INIT_HLIST_HEAD(&kvm->mask_notifier_list);
363 INIT_HLIST_HEAD(&kvm->irq_ack_notifier_list);
963#endif 364#endif
964 365
965#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 366#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
966 page = alloc_page(GFP_KERNEL | __GFP_ZERO); 367 page = alloc_page(GFP_KERNEL | __GFP_ZERO);
967 if (!page) { 368 if (!page) {
968 kfree(kvm); 369 r = -ENOMEM;
969 return ERR_PTR(-ENOMEM); 370 goto out_err;
970 } 371 }
971 kvm->coalesced_mmio_ring = 372 kvm->coalesced_mmio_ring =
972 (struct kvm_coalesced_mmio_ring *)page_address(page); 373 (struct kvm_coalesced_mmio_ring *)page_address(page);
@@ -974,15 +375,13 @@ static struct kvm *kvm_create_vm(void)
974 375
975#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER) 376#if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
976 { 377 {
977 int err;
978 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops; 378 kvm->mmu_notifier.ops = &kvm_mmu_notifier_ops;
979 err = mmu_notifier_register(&kvm->mmu_notifier, current->mm); 379 r = mmu_notifier_register(&kvm->mmu_notifier, current->mm);
980 if (err) { 380 if (r) {
981#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET 381#ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
982 put_page(page); 382 put_page(page);
983#endif 383#endif
984 kfree(kvm); 384 goto out_err;
985 return ERR_PTR(err);
986 } 385 }
987 } 386 }
988#endif 387#endif
@@ -1006,6 +405,12 @@ static struct kvm *kvm_create_vm(void)
1006#endif 405#endif
1007out: 406out:
1008 return kvm; 407 return kvm;
408
409out_err:
410 hardware_disable_all();
411out_err_nodisable:
412 kfree(kvm);
413 return ERR_PTR(r);
1009} 414}
1010 415
1011/* 416/*
@@ -1064,6 +469,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
1064 kvm_arch_flush_shadow(kvm); 469 kvm_arch_flush_shadow(kvm);
1065#endif 470#endif
1066 kvm_arch_destroy_vm(kvm); 471 kvm_arch_destroy_vm(kvm);
472 hardware_disable_all();
1067 mmdrop(mm); 473 mmdrop(mm);
1068} 474}
1069 475
@@ -1690,9 +1096,7 @@ void kvm_vcpu_block(struct kvm_vcpu *vcpu)
1690 if (signal_pending(current)) 1096 if (signal_pending(current))
1691 break; 1097 break;
1692 1098
1693 vcpu_put(vcpu);
1694 schedule(); 1099 schedule();
1695 vcpu_load(vcpu);
1696 } 1100 }
1697 1101
1698 finish_wait(&vcpu->wq, &wait); 1102 finish_wait(&vcpu->wq, &wait);
@@ -1706,6 +1110,21 @@ void kvm_resched(struct kvm_vcpu *vcpu)
1706} 1110}
1707EXPORT_SYMBOL_GPL(kvm_resched); 1111EXPORT_SYMBOL_GPL(kvm_resched);
1708 1112
1113void kvm_vcpu_on_spin(struct kvm_vcpu *vcpu)
1114{
1115 ktime_t expires;
1116 DEFINE_WAIT(wait);
1117
1118 prepare_to_wait(&vcpu->wq, &wait, TASK_INTERRUPTIBLE);
1119
1120 /* Sleep for 100 us, and hope lock-holder got scheduled */
1121 expires = ktime_add_ns(ktime_get(), 100000UL);
1122 schedule_hrtimeout(&expires, HRTIMER_MODE_ABS);
1123
1124 finish_wait(&vcpu->wq, &wait);
1125}
1126EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin);
1127
1709static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1128static int kvm_vcpu_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
1710{ 1129{
1711 struct kvm_vcpu *vcpu = vma->vm_file->private_data; 1130 struct kvm_vcpu *vcpu = vma->vm_file->private_data;
@@ -1829,88 +1248,6 @@ static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu *vcpu, sigset_t *sigset)
1829 return 0; 1248 return 0;
1830} 1249}
1831 1250
1832#ifdef __KVM_HAVE_MSIX
1833static int kvm_vm_ioctl_set_msix_nr(struct kvm *kvm,
1834 struct kvm_assigned_msix_nr *entry_nr)
1835{
1836 int r = 0;
1837 struct kvm_assigned_dev_kernel *adev;
1838
1839 mutex_lock(&kvm->lock);
1840
1841 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1842 entry_nr->assigned_dev_id);
1843 if (!adev) {
1844 r = -EINVAL;
1845 goto msix_nr_out;
1846 }
1847
1848 if (adev->entries_nr == 0) {
1849 adev->entries_nr = entry_nr->entry_nr;
1850 if (adev->entries_nr == 0 ||
1851 adev->entries_nr >= KVM_MAX_MSIX_PER_DEV) {
1852 r = -EINVAL;
1853 goto msix_nr_out;
1854 }
1855
1856 adev->host_msix_entries = kzalloc(sizeof(struct msix_entry) *
1857 entry_nr->entry_nr,
1858 GFP_KERNEL);
1859 if (!adev->host_msix_entries) {
1860 r = -ENOMEM;
1861 goto msix_nr_out;
1862 }
1863 adev->guest_msix_entries = kzalloc(
1864 sizeof(struct kvm_guest_msix_entry) *
1865 entry_nr->entry_nr, GFP_KERNEL);
1866 if (!adev->guest_msix_entries) {
1867 kfree(adev->host_msix_entries);
1868 r = -ENOMEM;
1869 goto msix_nr_out;
1870 }
1871 } else /* Not allowed set MSI-X number twice */
1872 r = -EINVAL;
1873msix_nr_out:
1874 mutex_unlock(&kvm->lock);
1875 return r;
1876}
1877
1878static int kvm_vm_ioctl_set_msix_entry(struct kvm *kvm,
1879 struct kvm_assigned_msix_entry *entry)
1880{
1881 int r = 0, i;
1882 struct kvm_assigned_dev_kernel *adev;
1883
1884 mutex_lock(&kvm->lock);
1885
1886 adev = kvm_find_assigned_dev(&kvm->arch.assigned_dev_head,
1887 entry->assigned_dev_id);
1888
1889 if (!adev) {
1890 r = -EINVAL;
1891 goto msix_entry_out;
1892 }
1893
1894 for (i = 0; i < adev->entries_nr; i++)
1895 if (adev->guest_msix_entries[i].vector == 0 ||
1896 adev->guest_msix_entries[i].entry == entry->entry) {
1897 adev->guest_msix_entries[i].entry = entry->entry;
1898 adev->guest_msix_entries[i].vector = entry->gsi;
1899 adev->host_msix_entries[i].entry = entry->entry;
1900 break;
1901 }
1902 if (i == adev->entries_nr) {
1903 r = -ENOSPC;
1904 goto msix_entry_out;
1905 }
1906
1907msix_entry_out:
1908 mutex_unlock(&kvm->lock);
1909
1910 return r;
1911}
1912#endif
1913
1914static long kvm_vcpu_ioctl(struct file *filp, 1251static long kvm_vcpu_ioctl(struct file *filp,
1915 unsigned int ioctl, unsigned long arg) 1252 unsigned int ioctl, unsigned long arg)
1916{ 1253{
@@ -2169,112 +1506,6 @@ static long kvm_vm_ioctl(struct file *filp,
2169 break; 1506 break;
2170 } 1507 }
2171#endif 1508#endif
2172#ifdef KVM_CAP_DEVICE_ASSIGNMENT
2173 case KVM_ASSIGN_PCI_DEVICE: {
2174 struct kvm_assigned_pci_dev assigned_dev;
2175
2176 r = -EFAULT;
2177 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2178 goto out;
2179 r = kvm_vm_ioctl_assign_device(kvm, &assigned_dev);
2180 if (r)
2181 goto out;
2182 break;
2183 }
2184 case KVM_ASSIGN_IRQ: {
2185 r = -EOPNOTSUPP;
2186 break;
2187 }
2188#ifdef KVM_CAP_ASSIGN_DEV_IRQ
2189 case KVM_ASSIGN_DEV_IRQ: {
2190 struct kvm_assigned_irq assigned_irq;
2191
2192 r = -EFAULT;
2193 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2194 goto out;
2195 r = kvm_vm_ioctl_assign_irq(kvm, &assigned_irq);
2196 if (r)
2197 goto out;
2198 break;
2199 }
2200 case KVM_DEASSIGN_DEV_IRQ: {
2201 struct kvm_assigned_irq assigned_irq;
2202
2203 r = -EFAULT;
2204 if (copy_from_user(&assigned_irq, argp, sizeof assigned_irq))
2205 goto out;
2206 r = kvm_vm_ioctl_deassign_dev_irq(kvm, &assigned_irq);
2207 if (r)
2208 goto out;
2209 break;
2210 }
2211#endif
2212#endif
2213#ifdef KVM_CAP_DEVICE_DEASSIGNMENT
2214 case KVM_DEASSIGN_PCI_DEVICE: {
2215 struct kvm_assigned_pci_dev assigned_dev;
2216
2217 r = -EFAULT;
2218 if (copy_from_user(&assigned_dev, argp, sizeof assigned_dev))
2219 goto out;
2220 r = kvm_vm_ioctl_deassign_device(kvm, &assigned_dev);
2221 if (r)
2222 goto out;
2223 break;
2224 }
2225#endif
2226#ifdef KVM_CAP_IRQ_ROUTING
2227 case KVM_SET_GSI_ROUTING: {
2228 struct kvm_irq_routing routing;
2229 struct kvm_irq_routing __user *urouting;
2230 struct kvm_irq_routing_entry *entries;
2231
2232 r = -EFAULT;
2233 if (copy_from_user(&routing, argp, sizeof(routing)))
2234 goto out;
2235 r = -EINVAL;
2236 if (routing.nr >= KVM_MAX_IRQ_ROUTES)
2237 goto out;
2238 if (routing.flags)
2239 goto out;
2240 r = -ENOMEM;
2241 entries = vmalloc(routing.nr * sizeof(*entries));
2242 if (!entries)
2243 goto out;
2244 r = -EFAULT;
2245 urouting = argp;
2246 if (copy_from_user(entries, urouting->entries,
2247 routing.nr * sizeof(*entries)))
2248 goto out_free_irq_routing;
2249 r = kvm_set_irq_routing(kvm, entries, routing.nr,
2250 routing.flags);
2251 out_free_irq_routing:
2252 vfree(entries);
2253 break;
2254 }
2255#endif /* KVM_CAP_IRQ_ROUTING */
2256#ifdef __KVM_HAVE_MSIX
2257 case KVM_ASSIGN_SET_MSIX_NR: {
2258 struct kvm_assigned_msix_nr entry_nr;
2259 r = -EFAULT;
2260 if (copy_from_user(&entry_nr, argp, sizeof entry_nr))
2261 goto out;
2262 r = kvm_vm_ioctl_set_msix_nr(kvm, &entry_nr);
2263 if (r)
2264 goto out;
2265 break;
2266 }
2267 case KVM_ASSIGN_SET_MSIX_ENTRY: {
2268 struct kvm_assigned_msix_entry entry;
2269 r = -EFAULT;
2270 if (copy_from_user(&entry, argp, sizeof entry))
2271 goto out;
2272 r = kvm_vm_ioctl_set_msix_entry(kvm, &entry);
2273 if (r)
2274 goto out;
2275 break;
2276 }
2277#endif
2278 case KVM_IRQFD: { 1509 case KVM_IRQFD: {
2279 struct kvm_irqfd data; 1510 struct kvm_irqfd data;
2280 1511
@@ -2306,11 +1537,59 @@ static long kvm_vm_ioctl(struct file *filp,
2306#endif 1537#endif
2307 default: 1538 default:
2308 r = kvm_arch_vm_ioctl(filp, ioctl, arg); 1539 r = kvm_arch_vm_ioctl(filp, ioctl, arg);
1540 if (r == -ENOTTY)
1541 r = kvm_vm_ioctl_assigned_device(kvm, ioctl, arg);
2309 } 1542 }
2310out: 1543out:
2311 return r; 1544 return r;
2312} 1545}
2313 1546
1547#ifdef CONFIG_COMPAT
1548struct compat_kvm_dirty_log {
1549 __u32 slot;
1550 __u32 padding1;
1551 union {
1552 compat_uptr_t dirty_bitmap; /* one bit per page */
1553 __u64 padding2;
1554 };
1555};
1556
1557static long kvm_vm_compat_ioctl(struct file *filp,
1558 unsigned int ioctl, unsigned long arg)
1559{
1560 struct kvm *kvm = filp->private_data;
1561 int r;
1562
1563 if (kvm->mm != current->mm)
1564 return -EIO;
1565 switch (ioctl) {
1566 case KVM_GET_DIRTY_LOG: {
1567 struct compat_kvm_dirty_log compat_log;
1568 struct kvm_dirty_log log;
1569
1570 r = -EFAULT;
1571 if (copy_from_user(&compat_log, (void __user *)arg,
1572 sizeof(compat_log)))
1573 goto out;
1574 log.slot = compat_log.slot;
1575 log.padding1 = compat_log.padding1;
1576 log.padding2 = compat_log.padding2;
1577 log.dirty_bitmap = compat_ptr(compat_log.dirty_bitmap);
1578
1579 r = kvm_vm_ioctl_get_dirty_log(kvm, &log);
1580 if (r)
1581 goto out;
1582 break;
1583 }
1584 default:
1585 r = kvm_vm_ioctl(filp, ioctl, arg);
1586 }
1587
1588out:
1589 return r;
1590}
1591#endif
1592
2314static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf) 1593static int kvm_vm_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
2315{ 1594{
2316 struct page *page[1]; 1595 struct page *page[1];
@@ -2345,7 +1624,9 @@ static int kvm_vm_mmap(struct file *file, struct vm_area_struct *vma)
2345static struct file_operations kvm_vm_fops = { 1624static struct file_operations kvm_vm_fops = {
2346 .release = kvm_vm_release, 1625 .release = kvm_vm_release,
2347 .unlocked_ioctl = kvm_vm_ioctl, 1626 .unlocked_ioctl = kvm_vm_ioctl,
2348 .compat_ioctl = kvm_vm_ioctl, 1627#ifdef CONFIG_COMPAT
1628 .compat_ioctl = kvm_vm_compat_ioctl,
1629#endif
2349 .mmap = kvm_vm_mmap, 1630 .mmap = kvm_vm_mmap,
2350}; 1631};
2351 1632
@@ -2373,6 +1654,7 @@ static long kvm_dev_ioctl_check_extension_generic(long arg)
2373#ifdef CONFIG_KVM_APIC_ARCHITECTURE 1654#ifdef CONFIG_KVM_APIC_ARCHITECTURE
2374 case KVM_CAP_SET_BOOT_CPU_ID: 1655 case KVM_CAP_SET_BOOT_CPU_ID:
2375#endif 1656#endif
1657 case KVM_CAP_INTERNAL_ERROR_DATA:
2376 return 1; 1658 return 1;
2377#ifdef CONFIG_HAVE_KVM_IRQCHIP 1659#ifdef CONFIG_HAVE_KVM_IRQCHIP
2378 case KVM_CAP_IRQ_ROUTING: 1660 case KVM_CAP_IRQ_ROUTING:
@@ -2443,11 +1725,21 @@ static struct miscdevice kvm_dev = {
2443static void hardware_enable(void *junk) 1725static void hardware_enable(void *junk)
2444{ 1726{
2445 int cpu = raw_smp_processor_id(); 1727 int cpu = raw_smp_processor_id();
1728 int r;
2446 1729
2447 if (cpumask_test_cpu(cpu, cpus_hardware_enabled)) 1730 if (cpumask_test_cpu(cpu, cpus_hardware_enabled))
2448 return; 1731 return;
1732
2449 cpumask_set_cpu(cpu, cpus_hardware_enabled); 1733 cpumask_set_cpu(cpu, cpus_hardware_enabled);
2450 kvm_arch_hardware_enable(NULL); 1734
1735 r = kvm_arch_hardware_enable(NULL);
1736
1737 if (r) {
1738 cpumask_clear_cpu(cpu, cpus_hardware_enabled);
1739 atomic_inc(&hardware_enable_failed);
1740 printk(KERN_INFO "kvm: enabling virtualization on "
1741 "CPU%d failed\n", cpu);
1742 }
2451} 1743}
2452 1744
2453static void hardware_disable(void *junk) 1745static void hardware_disable(void *junk)
@@ -2460,11 +1752,52 @@ static void hardware_disable(void *junk)
2460 kvm_arch_hardware_disable(NULL); 1752 kvm_arch_hardware_disable(NULL);
2461} 1753}
2462 1754
1755static void hardware_disable_all_nolock(void)
1756{
1757 BUG_ON(!kvm_usage_count);
1758
1759 kvm_usage_count--;
1760 if (!kvm_usage_count)
1761 on_each_cpu(hardware_disable, NULL, 1);
1762}
1763
1764static void hardware_disable_all(void)
1765{
1766 spin_lock(&kvm_lock);
1767 hardware_disable_all_nolock();
1768 spin_unlock(&kvm_lock);
1769}
1770
1771static int hardware_enable_all(void)
1772{
1773 int r = 0;
1774
1775 spin_lock(&kvm_lock);
1776
1777 kvm_usage_count++;
1778 if (kvm_usage_count == 1) {
1779 atomic_set(&hardware_enable_failed, 0);
1780 on_each_cpu(hardware_enable, NULL, 1);
1781
1782 if (atomic_read(&hardware_enable_failed)) {
1783 hardware_disable_all_nolock();
1784 r = -EBUSY;
1785 }
1786 }
1787
1788 spin_unlock(&kvm_lock);
1789
1790 return r;
1791}
1792
2463static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val, 1793static int kvm_cpu_hotplug(struct notifier_block *notifier, unsigned long val,
2464 void *v) 1794 void *v)
2465{ 1795{
2466 int cpu = (long)v; 1796 int cpu = (long)v;
2467 1797
1798 if (!kvm_usage_count)
1799 return NOTIFY_OK;
1800
2468 val &= ~CPU_TASKS_FROZEN; 1801 val &= ~CPU_TASKS_FROZEN;
2469 switch (val) { 1802 switch (val) {
2470 case CPU_DYING: 1803 case CPU_DYING:
@@ -2667,13 +2000,15 @@ static void kvm_exit_debug(void)
2667 2000
2668static int kvm_suspend(struct sys_device *dev, pm_message_t state) 2001static int kvm_suspend(struct sys_device *dev, pm_message_t state)
2669{ 2002{
2670 hardware_disable(NULL); 2003 if (kvm_usage_count)
2004 hardware_disable(NULL);
2671 return 0; 2005 return 0;
2672} 2006}
2673 2007
2674static int kvm_resume(struct sys_device *dev) 2008static int kvm_resume(struct sys_device *dev)
2675{ 2009{
2676 hardware_enable(NULL); 2010 if (kvm_usage_count)
2011 hardware_enable(NULL);
2677 return 0; 2012 return 0;
2678} 2013}
2679 2014
@@ -2748,7 +2083,6 @@ int kvm_init(void *opaque, unsigned int vcpu_size,
2748 goto out_free_1; 2083 goto out_free_1;
2749 } 2084 }
2750 2085
2751 on_each_cpu(hardware_enable, NULL, 1);
2752 r = register_cpu_notifier(&kvm_cpu_notifier); 2086 r = register_cpu_notifier(&kvm_cpu_notifier);
2753 if (r) 2087 if (r)
2754 goto out_free_2; 2088 goto out_free_2;
@@ -2798,7 +2132,6 @@ out_free_3:
2798 unregister_reboot_notifier(&kvm_reboot_notifier); 2132 unregister_reboot_notifier(&kvm_reboot_notifier);
2799 unregister_cpu_notifier(&kvm_cpu_notifier); 2133 unregister_cpu_notifier(&kvm_cpu_notifier);
2800out_free_2: 2134out_free_2:
2801 on_each_cpu(hardware_disable, NULL, 1);
2802out_free_1: 2135out_free_1:
2803 kvm_arch_hardware_unsetup(); 2136 kvm_arch_hardware_unsetup();
2804out_free_0a: 2137out_free_0a: